BrainBox: 134,201 Annotated Brain MRI Slices for Medical VLM Training

134,201 brain MRI slices are consolidated from multiple 2D and 3D medical imaging datasets into a standardized format. Created by liamchalcroft and last updated in September 2025, the dataset is designed for fine-tuning vision-language models on medical image localization tasks. Each image includes structured metadata and precise bounding box annotations for pathological findings.

Use Cases

• Fine-tuning vision-language models for medical image localization based on the bounding box annotations.
• Training models to identify pathological findings in brain MRI scans based on the provided annotations.
• Benchmarking model performance on a standardized, consolidated collection of medical imaging data.
• Developing multi-modal AI systems that process both medical images and structured metadata.

Strengths

• Contains 134,201 brain MRI slices, providing a substantial volume of data for model training.
• Includes precise bounding box annotations for pathological findings, which are essential for supervised localization tasks.
• Consolidates multiple source datasets into a standardized 2D format optimized for VLM training.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last updated 2025-09-14 13:05:25; freshness should be verified.
• The description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

Consolidated from multiple 2D and 3D medical imaging datasets.

Freshness

Last updated 2025-09-14 13:05:25.